Amphetamine dependence (AD) is an important public health problem, which has been linked to persistent executive and affective processing impairments. Specifically, amphetamine dependent individuals (ADI) demonstrate poor decision-making, particularly with respect to keeping track of new information and using this knowledge to guide future decisions. In addition, ADI have difficulties in appraising and regulating negative emotion, which has been linked to poor social functioning and higher likelihood of relapse. Given that emotion plays an important role in modulating and guiding decision-making, surprisingly little is known about how such dysfunctions interact in ADI, and how they may contribute to impairments in everyday functioning. Bayesian learning models provide a way to formally represent an individual's beliefs about the environment and the dynamic updating of those beliefs when new observations are made. Such computational framework can help to better delineate the potential impact of emotion on the dynamic representation of individuals' expectations about choice options, on the degree to which this cumulative knowledge is used to predict future outcomes, and on the cognitive strategies individuals use to select actions. The proposed study will use such Bayesian modeling approach, combined with event-related functional magnetic resonance imaging (fMRI), to assess the neurocognitive processes underlying a) potential deficits in the strategic and predictive processes guiding reward-based decision-making in ADI and b) ADI potential overrepresentation of or failure to integrate negative emotion into such decision-making. To do so, 40 ADI and 40 healthy comparison subjects (CS) will perform a gambling task under neutral and sad mood, and while undergoing fMRI. Several decision strategies and the associated Bayesian outcome predictions will be estimated and correlated with neural activity at the time of decision. These measures will be used to assess and quantify differences between ADI and CS 1) in the computational processes underlying reward-based decision (Aim1) and 2) in the computational processes underlying the integration of negative emotion into reward-based decision-making (Aim 2). The outcomes of this study will help to refine neurocognitive models of AD by delineating the computational processes that go awry in ADI and will help identify more precise neurocognitive predictors of executive and affective dysfunction an ADI.